Modeling biodiversity benchmarks in variable environments

Effective environmental assessment and management requires quantifiable biodiversity targets. Biodiversity benchmarks define these targets by focusing on specific biodiversity metrics, such as species richness. However, setting fixed targets can be challenging because many biodiversity metrics are highly variable, both spatially and temporally. We present a multivariate, hierarchical Bayesian method to estimate biodiversity benchmarks based on the species richness and cover of native terrestrial vegetation growth forms. This approach uses existing data to quantify the empirical distributions of species richness and cover within growth forms, and we use the upper quantiles of these distributions to estimate contemporary, “best‐on‐offer” biodiversity benchmarks. Importantly, we allow benchmarks to differ among vegetation types, regions, and seasons, and with changes in recent rainfall. We apply our method to data collected over 30 yr at ~35,000 floristic plots in southeastern Australia. Our estimated benchmarks were broadly consistent with existing expert‐elicited benchmarks, available for a small subset of vegetation types. However, in comparison with expert‐elicited benchmarks, our data‐driven approach is transparent, repeatable, and updatable; accommodates important spatial and temporal variation; aligns modeled benchmarks directly with field data and the concept of best‐on‐offer benchmarks; and, where many benchmarks are required, is likely to be more efficient. Our approach is general and could be used broadly to estimate biodiversity targets from existing data in highly variable environments, which is especially relevant given rapid changes in global environmental conditions. Biodiversity benchmarks (hereafter, benchmarks) are routinely used in a range of assessment and management applications as the quantitative estimates of desirable biodiversity states (e.g., restoration ecology [Hobbs and Harris 2001], offsetting schemes [Bull et al. 2014a, b]). Despite debate on how best to apply these benchmarks (Suding 2011, Maron et al. 2012), it is agreed that natural resource managers require transparent and repeatable methods to quantify desirable and undesirable biodiversity states (Oliver et al. 2002, 2007, Quétier and Lavorel 2011, Pardo et al. 2012, Bull et al. 2014a, b). Deviations from benchmarks provide an estimate of the quality of a site and indicate the potential for improvements in biodiversity at a site (Sinclair et al. 2002). Benchmarks often represent quantitative estimates of “historical” (e.g., pre‐intensive agriculture, presettlement;